An Evidence Based Earthquake Detector using Twitter

This paper presents a notification system to identify earthquakes from firsthand reports published on Twitter. Tweets from target regions in Australia and New Zealand are checked for earthquake keyword frequency bursts and then processed to identify evidence of an earthquake. The benefit of our earthquake detector is that it relies on evidence of firsthand ‘felt’ reports from Twitter, provides an indication of the earthquake intensity and will be the trigger for further classification of Tweets for impact analysis. We describe how the detector has been incrementally improved, most notably by the introduction of a text classifier. During its initial five months of operation the system has generated 49 notifications of which 29 related to real earthquake events.